This cross-sectional study applied latent profile analysis to symptom data from 942 lung cancer survivors treated at four hospitals in Shanghai. Participants had either completed surgical treatment or received at least one course of initial antitumour therapy and were in a stable follow-up phase or treatment interval. The sample had a mean age of 64.43 ± 10.70 years and was 65.10% male.
Three latent symptom subgroups emerged from the analysis: a low-symptom group (n = 488, 51.80%), a moderate-symptom group (n = 364, 38.64%) and a high-symptom group (n = 90, 9.55%). These profiles reflect heterogeneity in symptom burden among survivors and provide a data-driven basis for stratifying patients by overall symptom severity patterns.
Symptom networks were derived using partial correlation methods to assess the structure and interconnectedness of symptoms within each subgroup. Network density was descriptively greater in the high-symptom group compared with the low-symptom group (0.468 vs 0.175), indicating that symptoms were more tightly interrelated when overall burden was higher.
Centrality analyses identified different central symptoms across subgroups. In the low-symptom group, cough was the central symptom. In the moderate-symptom group, vomiting emerged as central. In the high-symptom group, distress was the central node. These differences suggest that the most influential symptoms vary with overall symptom burden and that targeted interventions might differ by subgroup.
The investigators applied LASSO variable selection and collinearity assessment to generate a candidate set of predictors. After selection, 22 predictors were entered into a multivariable multinomial logistic regression model to examine factors associated with subgroup membership.
Key associations reported in the adjusted model included:
Surgical treatment with adjuvant therapy was associated with higher odds of belonging to the moderate-symptom group versus the low-symptom group (adjusted OR = 4.054, 95% CI 2.094 to 7.850).
Better exercise capacity, operationalized as a 6-minute walk distance ≥ 450 m, was associated with substantially lower odds of membership in the high-symptom group (adjusted OR = 0.101, 95% CI 0.027 to 0.372).
The final multivariable model had a Nagelkerke pseudo-R² of 0.522, indicating that the included predictors accounted for a meaningful portion of variance in subgroup membership.
The findings demonstrate heterogeneity in both symptom severity profiles and underlying symptom network structures among lung cancer survivors. The higher network density observed in the high-symptom group implies greater symptom interdependence when burden is high; identifying central symptoms in each subgroup—cough, vomiting, and distress—may help prioritize targets for intervention.
Integrating latent profile analysis with symptom network analysis offers a framework for stratified symptom assessment and individualized symptom management in survivorship care. For example, survivors classified in a high-symptom profile might benefit from interventions that address psychological distress as a central driver, whereas survivors in other profiles may require different primary symptom targets.
The associations between treatment modality (surgery with adjuvant therapy) and elevated odds of moderate symptom burden, and between preserved exercise capacity and reduced likelihood of high symptom burden, suggest potential clinical levers for identifying at-risk patients and for intervention planning.
This study was cross-sectional and conducted across four hospitals in Shanghai: a national thoracic oncology centre, two tertiary general hospitals and one regional general hospital. The sample included 942 lung cancer survivors who were in stable follow-up or a treatment interval after completing surgery or at least one course of initial antitumour therapy. Mean age was 64.43 years (SD 10.70); 65.10% of participants were male.
Primary outcomes were latent symptom subgroups identified by latent profile analysis and symptom network characteristics derived from partial correlation networks. Secondary outcomes were socio-demographic, clinical, functional and psychosocial factors associated with subgroup membership, assessed via LASSO selection and multivariable multinomial logistic regression.
As reported, the study design was cross-sectional, which limits causal inference between predictors and subgroup membership. The source reports the trial registration number as MR-31-24-027806 and provides a registry link. Specific measures, symptom instruments, and certain analytic details beyond those summarized were not reported in the source text provided here.
Conclusions
Symptom burden among lung cancer survivors in this sample is heterogeneous in both severity and network structure. Combining latent profile analysis with symptom network analysis identified distinct subgroups and differing central symptoms, and highlighted clinical predictors associated with subgroup membership. This combined approach may support more tailored symptom assessment and individualized management strategies in survivorship care.